CN106777594A - A kind of self-learning method of the operation of rolling Mathematical Modeling based on continuous curve surface - Google Patents

Abstract

A kind of self-learning method of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, has the feature that, comprises the following steps：Step one, sets up form, for the corresponding Model Self-Learning coefficient of each characteristic point in memory space；Step 2, before belt steel rolling, intends interpolation calculation and obtains the corresponding Model Self-Learning coefficient of actual rolling operating point according to continuous curve surface；And step 3, after the completion of belt steel rolling, Model Self-Learning coefficient form is updated according to weight coefficient matrix, will pass through the setting accuracy that study improves constantly rolling model, for the rolling of follow-up strip.The present invention is other with the continuous curve surface substitutable layer that characteristic point is characterized, plan interpolation is carried out to the self study coefficient in each characteristic point in space using continuous function, obtain the continuous self study coefficient smooth surface in any point, the continuous treatment of implementation model self study coefficient, and all specifications are quickly extended to by part specification, the final setting accuracy for improving operation of rolling Mathematical Modeling.

Description

A kind of self-learning method of the operation of rolling Mathematical Modeling based on continuous curve surface

Technical field

The present invention relates to metallurgical industry operation of rolling Mathematical Modeling, more particularly to a kind of operation of rolling based on continuous curve surface The self-learning method of Mathematical Modeling.

Background technology

Operation of rolling Mathematical Modeling is set up on rolling therory basis mostly, due to calculating speed and application performance Limitation, what is used on rolling line at present is the simplified model obtained on the basis of theory.Due to being particularly deformed area on rolling line Interior some events and phenomenon not yet obtain perfect theoretical explanation, such as the change of friction condition, metal are becoming in deformed area Flowing law in shape area etc.；Some conventional hypothesis have differences with actual conditions, such as roll flattening shape is assumed, plane becomes Shape is assumed；The convective region of the water-cooled mechanism in cooling procedure, nuclear boiling area, film boiling area, small liquid accumulation regions are guessed；These Problem limits the computational accuracy and stability of operation of rolling Mathematical Modeling.

The use of the other data of layer provides point of penetration to improve the computational accuracy of operation of rolling Mathematical Modeling.At present no matter state Outer main flow Mathematical Modeling or the Mathematical Modeling of domestic independent development, using " mechanism model+layer does not divide+self adaptation " mechanism To build model.Layer did not divided and is slightly then unprofitable to improve model computational accuracy, crosses detailed rules and regulations and improves debugging difficulty, reduced model Performance, needs therebetween a kind of balance.

At present, operation of rolling mathematical modulo is built using " mechanism model+layer does not divide+self adaptation " mechanism both at home and abroad Type, is not divided by layer and is grouped rolling operating mode, and one " layer is other " represents one by continuous specification variable (such as thickness, width Degree) determine area of space, and correspondence one Model Self-Learning coefficient.After the completion of per winding steel rolling, model according to operating mode not It is disconnected to update the other self study coefficient of respective layer.Due to rolling the fluctuation of operating mode and the nebula distribution characteristics of procedure parameter, learn by oneself Practise coefficient renewal concentrate on it is a part of often roll the corresponding layer of specification and do not go up, distributed pole is uneven, is shown in actual production Come problem be：1. model specification precision is not updated the frequency and is influenceed by current layer, update the more other model accuracy of layer compared with Height, and update that the other model accuracy of less layer is relatively low, model specification precision is unstable；When 2. changing specification rolling, due to difference The other Model Self-Learning coefficient of layer is discontinuous, jump is big, causes model specification precision poor, influence belt steel product quality index Lifting.It can be seen that, there is design defect in original rolling Mathematical Modeling self-learning method：1. being layered does not carry out model parameter self-study Practise, although model accuracy when can ensure that same specification is rolled by " local linearization " on regional area, but cause different layers The problems such as other self study coefficient is mutually unrelated, discontinuous, jump is big；2. a belt steel rolling only updates a layer not, layer The expansion speed of other data is slower；When 3. running into new spec rolling, the extension of the other data of layer is a problem for being difficult to avoid.

The content of the invention

The present invention is carried out to solve the above problems, it is therefore intended that provided using continuous curve surface substitutable layer not, real The rapid specification of existing Model Self-Learning coefficient is expanded, and no matter is under what circumstances realized, including change layer do not roll, new spec and new Kind trial-production etc., model will keep a kind of operation of rolling based on continuous curve surface of setting accuracy higher and adaptive ability The self-learning method of Mathematical Modeling.

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, with such Feature, comprises the following steps：

Step one, sets up form, for the corresponding Model Self-Learning coefficient of each characteristic point in memory space；

Step 2, before belt steel rolling, intends interpolation calculation and obtains the corresponding model of actual rolling operating point according to continuous curve surface Self study coefficient；And

Step 3, after the completion of belt steel rolling, updates Model Self-Learning coefficient form, will pass through according to weight coefficient matrix Study improves constantly the setting accuracy of rolling model, for the rolling of follow-up strip.

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, in the space determined by continuous specification variable, multiple characteristic points are evenly arranged, by the principle that grid is rule Take a little.

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, the key-value pair of form answers the sequence number of characteristic point, each characteristic point one Model Self-Learning coefficient of correspondence.

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, step 2, it is necessary to first obtain Model Self-Learning coefficient before belt steel rolling, is then used further to operation of rolling mathematics The setup algorithm of model, interpolation method is intended using continuous function, and the model in each characteristic point in specifications parameter space is learnt by oneself Practising coefficient carries out plan interpolation, obtains the self study coefficient function that space any point is continuous and can lead, self study coefficient function It is the function of locus coordinate, the precision of Model Self-Learning coefficient is accurate to each position from the other region of original layer and sit Mark.

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, the formula of continuous function plan interpolation method is：

Wherein,

h_x=max { x_i+1-x_i} (8)

h_y=max { y_j+1-y_j} (9)

There is property：

T in formula (1)^*(x, y) represents the Model Self-Learning coefficient of space optional position, if these coefficients connected Then to form a smooth surface；X, y represent original division layer all kinds of continuous variables not, x_i, y_jRepresentative feature point is in sky Between position coordinates on each change in coordinate axis direction；T_i,jRepresent the Model Self-Learning coefficient in each characteristic point.

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, the formula of continuous function plan interpolation method is：

There is property：

T in formula (11)^*(x, y, z) represents the Model Self-Learning coefficient of space optional position, if by these coefficients Link up then one smooth surface of formation；X, y, z represent original division layer all kinds of continuous variables not,

x_i,y_j,z_kPosition coordinates of the representative feature point on each change in coordinate axis direction in space；T_i,j,kRepresent in each characteristic point Model Self-Learning coefficient.

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, the storage of Model Self-Learning coefficient, using rule be：

1. each characteristic point carries out serial number by change in coordinate axis direction, and classified variable inherently classification number, both groups Close the index key assignments for forming storage form；

2. the use of Model Self-Learning coefficient, the combined index for first being formed according to the numbering of classified variable and characteristic point takes out Self study coefficient in each characteristic point, the corresponding Model Self-Learning of current rolling operating mode (x, y) is calculated according still further to formula (1) Coefficient T^*(x,y)。

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, the storage of Model Self-Learning coefficient, using rule be：

1. each characteristic point carries out serial number by change in coordinate axis direction, and classified variable inherently classification number, both groups Close the index key assignments for forming storage form；

2. the use of Model Self-Learning coefficient, the combined index for first being formed according to the numbering of classified variable and characteristic point takes out Self study coefficient in each characteristic point, the corresponding model of current rolling operating mode (x, y, z) is calculated certainly according still further to formula (11) Learning coefficient T^*(x,y,z)。

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, after the completion of belt steel rolling, Model Self-Learning coefficient is updated.Actual rolling operating mode (x is calculated first_act, y_act),

Weight coefficient q in each characteristic point of space_{I, j}：

Then the self study coefficient in renewal space in each characteristic point, updates according to the following formula for multiplying inquiry learning coefficient：

Wherein α is smoothing factor, and span is 0 ＜ α ＜ 1, K_actAfter measured value and deduction Model Self-Learning coefficient The ratio of calculated value；

For additivity learning coefficient (such as temperature model), update according to the following formula：

Wherein α is smoothing factor, C_actIt is measured value and the difference for deducting the calculated value after Model Self-Learning coefficient, works as reality Closer to any one characteristic point, then the corresponding weight coefficient of this feature point is relatively bigger for the corresponding operating point of rolling operating mode, its Obtain the ratio for updating bigger.

The self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface that the present invention is provided, also with so Feature：Wherein, after the completion of belt steel rolling, Model Self-Learning coefficient is updated.Actual rolling operating mode (x is calculated first_act, y_act,z_act),

Weight coefficient q in each characteristic point of space_{I, j, k}：

Then the self study coefficient in renewal space in each characteristic point, updates according to the following formula for multiplying inquiry learning coefficient：

Wherein α is smoothing factor, and span is 0 ＜ α ＜ 1, K_actAfter measured value and deduction Model Self-Learning coefficient The ratio of calculated value；

For additivity learning coefficient (such as temperature model), update according to the following formula：

Wherein α is smoothing factor, C_actIt is measured value and the difference for deducting the calculated value after Model Self-Learning coefficient, works as reality Closer to any one characteristic point, then the corresponding weight coefficient of this feature point is relatively bigger for the corresponding operating point of rolling operating mode, its Obtain the ratio for updating bigger.

The effect of invention and effect

According to a kind of self-learning method of the operation of rolling Mathematical Modeling based on continuous curve surface involved in the present invention, feature is used The continuous curve surface substitutable layer that point is characterized not, carries out intending inserting using continuous function to the self study coefficient in each characteristic point in space Value, obtains the self study coefficient smooth surface that spatially any point is continuous and can lead, and implementation model self study coefficient is by portion The other rolling of divider compartment is quickly extended to all specifications, and different size self study coefficient continuous treatment, finally may be used To improve the setting accuracy of operation of rolling Mathematical Modeling.

Brief description of the drawings

Fig. 1 is the other form of rolling force model self study coefficient corresponding layer of the present invention in embodiment.

Fig. 2 is the comparing figure of in embodiment " layer is not divided " of the invention.

Fig. 3 is the comparing figure of in embodiment " characteristic point+plan interpolation " of the invention.

Fig. 4 is the schematic diagram that Model Self-Learning coefficient of the present invention in embodiment stores form.

Fig. 5 is that rate of deformation of the present invention in embodiment is 30s-1, and curved surface when rolling temperature is 875 DEG C intends interpolation Result figure.

Fig. 6 is the signal of the nearer corresponding weight coefficient of four characteristic points of distance objective point of the present invention in embodiment Figure.

Fig. 7 is that the variable quantity (new coefficient-old coefficient) of self study coefficient in each grid in embodiment of the invention shows It is intended to.

Fig. 8 is the result figure that floor data with character pair point of the present invention in embodiment intend interpolation.

Specific embodiment

Referring to the drawings and embodiment is to a kind of operation of rolling mathematical modulo based on continuous curve surface involved in the present invention The self-learning method of type is explained in detail.

A kind of self-learning method of the operation of rolling Mathematical Modeling based on continuous curve surface, it is characterised in that it includes following Step：

Step one：Form is set up, for the corresponding Model Self-Learning coefficient of each characteristic point in memory space, into step Two.

Traditional rolling model self-learning method is needed to carry out layer not divide, and Model Self-Learning coefficient is stored in the other table of layer Lattice, a layer Dui Ying not a Model Self-Learning coefficient.By taking rolling force model as an example, the other Table Design of layer is as shown in figure 1, form It is other layer to be divided by " steel grade × frame × finish to gauge thickness × finish to gauge width ".Wherein, " steel grade × frame " is classified variable, and " eventually Roll thickness × finish to gauge width " it is then continuous variable, continuous variable needs to carry out stepping according to its criteria for classification, so as to obtain each Layer alias 1 on parametric direction, 2 ....

This patent proposes the concept of characteristic point, in the space determined by continuous specification variable, is evenly arranged multiple features Point, as shown in Figures 2 and 3.Stored for the ease of computer table, carried out according to the principle of " smooth vertical and horizontal " (being rule by grid) Take a little, but on same axis (such as finish to gauge thickness, finish to gauge width), the interval for taking a little can be the same or different, according to specification The characteristics of parameter, determines.

For storage model self study coefficient, it is necessary to set up a form, the key-value pair of form answers the sequence number of characteristic point, with It is easy to the storage of computer to operate.From unlike former method, be now each characteristic point correspondence one Model Self-Learning system Number, and be in the past that a layer Dui Ying not a self study coefficient.

Step 2：Before belt steel rolling, interpolation calculation is intended according to continuous curve surface and obtains the corresponding model of actual rolling operating point Self study coefficient, into step 3.

, it is necessary to first obtain the self study coefficient of model before belt steel rolling, setting for operation of rolling Mathematical Modeling is then used further to Devise a stratagem is calculated, and is given below and how to be obtained Model Self-Learning coefficient by continuous curve surface plan interpolation.

Basic ideas are to intend interpolation method using continuous function, to the self study in each characteristic point in specifications parameter space Coefficient carries out plan interpolation, obtains the self study coefficient function (smooth surface) that space any point is continuous and can lead；The function It is the function of locus coordinate, the precision of self study coefficient is accurate to each position coordinates from the other region of original layer, solves The other self study coefficient of existing different layers of having determined is discontinuous, big problem of jumping.

Find functional form that is appropriate continuous and can leading, using the continuous function in each characteristic point in space from Learning coefficient, carries out plan interpolation calculation, is superimposed the continuous function in all characteristic points, obtains spatially optional position and continuously may be used The self study coefficient function led, is shown below：

Wherein,

h_x=max { x_i+1-x_i} (8)

h_y=max { y_j+1-y_j} (9)

There is property：

T in formula (1)^*(x, y) represents the Model Self-Learning coefficient of space optional position, if these coefficients connected Then to form a smooth surface；X, y represent original division layer all kinds of continuous variables not, the generation if to rolling force model Table finish to gauge thickness, finish to gauge width；x_i, y_jPosition coordinates of the representative feature point on each change in coordinate axis direction in space；T_i,jRepresent each Self study coefficient in characteristic point, the value will be used for Computer Storage and update.

Formula (1) is that, in the presence of 2 situations of variable, such as to 3 variables, then formula (1) can be write as：

There is property：

The storage of Model Self-Learning coefficient, use rule：1. use and calculated with the other form similar mode of existing layer Machine is stored, and each characteristic point is carried out into serial number (i, j) or (i, j, k) by change in coordinate axis direction, and classified variable is inherently Classification number, therefore can be formed and store form with as the other data class of existing layer；2. the use of self study coefficient, first becomes according to classification Amount numbers the self study coefficient T that the combined index to be formed is taken out in each characteristic point with characteristic point_i,jOr T_i,j,k, according still further to formula Or (11) calculate the corresponding self study coefficient T of current rolling operating mode (x, y) or (x, y, z) (1)^*(x, y) or T^*(x,y,z)。

Step 3：After the completion of belt steel rolling, Model Self-Learning coefficient form is updated according to weight coefficient matrix, to lead to The setting accuracy that study improves constantly rolling model is crossed, for the rolling of follow-up strip.

, it is necessary to be updated to Model Self-Learning coefficient after the completion of belt steel rolling, to improve constantly the setting essence of model Degree.Weight coefficient q of actual rolling operating mode (x, y) or (x, y, z) in each characteristic point of space is calculated first_{I, j}Or q_{I, j, k}：

Or

Then update the self study coefficient in each characteristic point in space, for multiply inquiry learning coefficient (such as rolling force model, Deformation resistance model etc.) update according to the following formula：

Or

Wherein α is smoothing factor, and span is 0 ＜ α ＜ 1；K_actIt is the meter after measured value and deduction model learning coefficient The ratio of calculation value；

For additivity learning coefficient (such as temperature model), update according to the following formula：

Or

Wherein α is smoothing factor, C_actIt is measured value and the difference for deducting the calculated value after Model Self-Learning coefficient.

When the actual corresponding operating point of rolling operating mode is closer to certain characteristic point, then the corresponding weight coefficient phase of this feature point To bigger, it is bigger that it obtains the ratio for updating.The self study coefficient in each characteristic point of space is updated by constantly study, More and more accurate self study coefficient continuous curve surface is progressively obtained, so as to improve constantly the setting essence of operation of rolling Mathematical Modeling Degree.

Below by taking domestic certain hot continuous rolling production line mm finishing mill unit deformation resistance model as an example.

It is other layer to be divided by " steel grade × frame × rate of deformation × rolling temperature ".Wherein, " steel grade × frame " is classification change Amount, and " rate of deformation × rolling temperature " is then continuous variable.Choose steel grade code SGF=9, frame F2 deformation resistance models pair Illustrated as a example by the self study coefficient data answered.First, in rolling, operating mode --- rate of deformation is 30s for calculating^-1, rolling temperature For 875 DEG C when Model Self-Learning coefficient；Then, according to " actual measurement " resistance of deformation (being obtained come inverse by surveying roll-force) come Update Model Self-Learning coefficient form.

Step one, form is set up, for the corresponding Model Self-Learning coefficient of each characteristic point in memory space；

Rate of deformation and deformation extent be divide into 20 grades, the self study coefficient form of deformation resistance model is set up such as Shown in Fig. 4, in one characteristic point of each grid center arrangement, the form shown in Fig. 4 is used for storing the corresponding model of each characteristic point Self study coefficient.Each characteristic point is carried out into serial number (i, j) by change in coordinate axis direction, i then represents abscissa line rate of deformation Grade, j then represents the grade of coordinate longitudinal axis rolling temperature, so as to facilitate computer to be indexed.

Before step 2, belt steel rolling, interpolation calculation is intended according to continuous curve surface and obtains the corresponding model of actual rolling operating point Self study coefficient；

Using continuous function to the self study coefficient in each characteristic point in space, plan interpolation calculation is carried out, superposition is all Continuous function in characteristic point, obtains the self study coefficient function that spatially optional position can continuously lead, and is shown below：

h_x=max{x_i+1-x_i} (26)

h_y=max { y_j+1-y_j} (27)

Wherein, T^*(x, y) represents the Model Self-Learning coefficient of space optional position, the shape if these coefficients linked up Into a smooth surface；X, y represent original division layer all kinds of continuous variables not, and rate of deformation and rolling temperature are represented here Degree.

According to actual rolling, operating mode --- rate of deformation is 30s^-1, rolling temperature is 875 DEG C, i.e., x takes 30, y and takes 875；x_i, y_jRepresent position coordinates of each characteristic point in the reference axis of space two；T_i,jRepresent the self study coefficient in each characteristic point；M The number of degrees 20 of rate of deformation is then taken, N then takes the number of degrees 20 of rolling temperature.

By intending interpolation calculation, operating mode --- rate of deformation is 30s can to obtain rolling^-1, when rolling temperature is 875 DEG C Self study coefficient T^*(30,875)=1.367, it is as shown in Figure 5 that curved surface intends interpolation situation.

After the completion of step 3, belt steel rolling, Model Self-Learning coefficient form is updated according to weight coefficient matrix；

, it is necessary to be updated to Model Self-Learning coefficient after the completion of belt steel rolling, to improve constantly the setting essence of model Degree.The weight coefficient q in each characteristic point of space is calculated first_i,j：

Then the self study coefficient in each characteristic point in space is updated, for deformation resistance model (multiplying inquiry learning coefficient) Update according to the following formula：

Wherein α is smoothing factor, and span is 0 ＜ α ＜ 1, and α=0.3 is taken here；K_actFor measured value learns with deduction The ratio of calculated value, takes K here after model after coefficient_act=1.6；

Because in rolling, operating mode --- rate of deformation is 30s^-1, when rolling temperature is 875 DEG C, Model Self-Learning coefficient form In be respectively from rolling nearest four characteristic points of operating point：Rate of deformation class 6 and rolling temperature grade 14, rate of deformation etc. Level 7 and rolling temperature grade 14, rate of deformation grade are 6 and rolling temperature grade 15, rate of deformation grade 7 and rolling temperature etc. Level 15, so the corresponding weight coefficient q of this four characteristic points_{I, j}It is larger, q_6,14=q_6,15=04152, q_7,14=q_7,15= 0.0833, as shown in Figure 6.Therefore, self study coefficient update in this four characteristic points significantly (i.e. Model Self-Learning coefficient Variable quantity is larger), as shown in Figure 7.Fig. 7 shows the situation of change of the Model Self-Learning coefficient on each grid (after renewal Self study coefficient before self study coefficient-renewal).

It is assumed below to embodiment as above, initial Model Self-Learning coefficient table value is not changed, but change is rolled Operating mode point data (i.e. the value of rate of deformation and rolling temperature) processed, each characteristic point of interpolation calculation space is intended by continuous curve surface Corresponding self study coefficient, and contrasted with original self study coefficient table value, its result is as shown in Figure 7.According to continuous Curved surface intends the result that interpolation calculation is obtained, and with original self study coefficient form Data Comparison, obtains its mean absolute error (MAE) it is only that 0.115% root-mean-square error (RMSE) is only 0.0023, this also demonstrates continuous function and intends interpolation calculation from side Error very little, precision it is very high.

Above-described embodiment shows：The patent can not only solve the other Model Self-Learning coefficient of original adjacent layer jump, Discontinuous problem, has further the advantage that：1. once rolling updates all of characteristic point, so model adapts to product specification and opens up The speed of exhibition is quickly；2. self study coefficient is used, updates the weight with each characteristic point, root all in accordance with the calculating of locus coordinate The annexation of present operating point and each characteristic point of space is determined according to weight, use and the renewal of self study coefficient have intelligence Energy property, therefore without manual intervention self study coefficient.

This patent method is mainly by proposing a kind of new operation of rolling Mathematical Modeling based on continuous curve surface plan interpolation Self-learning method, so as to improve setting for operation of rolling Mathematical Modeling (such as temperature model, rolling force model, deformation resistance model) Precision is determined, so as to improve the rolling stability of rolling process and every belt steel product Mass accuracy (such as thickness control accuracy, temperature Control accuracy, Strip Shape Control precision etc.).

The effect of embodiment and effect

A kind of self-learning method of the operation of rolling Mathematical Modeling based on continuous curve surface according to involved by the present embodiment, with spy Levy a continuous curve surface substitutable layer for characterizing not, the self study coefficient in each characteristic point in space is intended using continuous function Interpolation, obtains the self study coefficient smooth surface that spatially any point is continuous and can lead, implementation model self study coefficient by The other rolling of part specification layer is quickly extended to all specifications, and different size self study coefficient continuous treatment, finally The setting accuracy of operation of rolling Mathematical Modeling can be improved.

Above-mentioned implementation method is preferred case of the invention, is not intended to limit protection scope of the present invention.

Claims (10)

1. a kind of self-learning method of the operation of rolling Mathematical Modeling based on continuous curve surface, it is characterised in that it includes following step Suddenly：

Step one, sets up form, for the corresponding Model Self-Learning coefficient of each characteristic point in memory space；

Step 2, before belt steel rolling, intends interpolation calculation and obtains the corresponding model self-study of actual rolling operating point according to continuous curve surface Practise coefficient；And

Step 3, after the completion of belt steel rolling, updates Model Self-Learning coefficient form, will pass through study according to weight coefficient matrix The setting accuracy of rolling model is improved constantly, for the rolling of follow-up strip.

2. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 1, it is special Levy and be：

Wherein, in the space determined by continuous specification variable, multiple characteristic points are evenly arranged, by the principle that grid is rule Take a little.

3. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 1, it is special Levy and be：

Wherein, the key-value pair of the form answers the sequence number of the characteristic point,

Each one described Model Self-Learning coefficient of the characteristic point correspondence.

4. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 1, it is special Levy and be：

Wherein, the step 2, it is necessary to first obtain the Model Self-Learning coefficient before the belt steel rolling, is then used further to roll The setup algorithm of process mathematical model processed,

Interpolation method is intended using continuous function, to the Model Self-Learning system in each described characteristic point in specifications parameter space Number carries out plan interpolation, obtains the self study coefficient function that space any point is continuous and can lead,

The self study coefficient function is the function of locus coordinate, makes the precision of the Model Self-Learning coefficient from original The other region of layer is accurate to each position coordinates.

5. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 4, it is special Levy and be：

Wherein, the formula of the continuous function plan interpolation method is：

Wherein,

$\mathrm{\φ}\left(x\right)=\sqrt{{c_x}^2+x^2}---\left(4\right)$

$\mathrm{\φ}\left(y\right)=\sqrt{{c_y}^2+y^2}---\left(5\right)$

$c_x=0.1h_x^{\frac{1}{3}}---\left(6\right)$

$c_y=0.1h_y^{\frac{1}{3}}---\left(7\right)$

h_x=max (x_i+1-x_i) (8)

h_y=max { y_i+1-y_j) (9)

There is property：

T in formula (1)^*(x, y) represents the Model Self-Learning coefficient of space optional position, if these coefficients linked up Form a smooth surface；X, y represent original division layer all kinds of continuous variables not,

x_i, y_jPosition coordinates of the representative feature point on each change in coordinate axis direction in space；T_i,jRepresent the institute in each described characteristic point State Model Self-Learning coefficient.

6. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 4, it is special Levy and be：

Wherein, the formula of the continuous function plan interpolation method is：

There is property：

T in the formula (11)^*(x, y, z) represents the Model Self-Learning coefficient of space optional position, if these coefficients connected Get up and then form a smooth surface；X, y, z represent original division layer all kinds of continuous variables not,

x_i,y_j,z_kPosition coordinates of the representative feature point on each change in coordinate axis direction in space；T_i,j,kRepresent in each described characteristic point The Model Self-Learning coefficient.

7. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 5, it is special Levy and be：

Wherein, the storage of the Model Self-Learning coefficient, using rule be：

1. each described characteristic point carries out serial number by change in coordinate axis direction, and classified variable inherently classification number, both groups Close the index key assignments for forming storage form；

2. the use of the Model Self-Learning coefficient, the combined index for first being formed with the numbering of the characteristic point according to classified variable The self study coefficient in each characteristic point is taken out, the corresponding institute of current rolling operating mode (x, y) is calculated according still further to the formula (1) State Model Self-Learning coefficient T^*(x,y)。

8. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 6, it is special Levy and be：

Wherein, the storage of the Model Self-Learning coefficient, using rule be：

1. each described characteristic point carries out serial number by change in coordinate axis direction, and classified variable inherently classification number, both groups Close the index key assignments for forming storage form；

2. the use of the Model Self-Learning coefficient, the combined index for first being formed with the numbering of the characteristic point according to classified variable The self study coefficient in each characteristic point is taken out, current rolling operating mode (x, y, z) correspondence is calculated according still further to the formula (11) The Model Self-Learning coefficient T^*(x,y,z)。

9. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 4, it is special Levy and be：

Wherein, after the completion of the belt steel rolling, the Model Self-Learning coefficient is updated.Actual rolling operating mode is calculated first (x_act,y_act),

Weight coefficient q in space in each described characteristic point_{I, j}：

Then the self study coefficient in renewal space in each characteristic point, updates according to the following formula for multiplying inquiry learning coefficient：

$T_{i,j}^{new}=T_{i,j}+q_{i,j}\mathrm{\α}(K_{act}-T_{i,j})---\left(14\right)$

Wherein α is smoothing factor, and span is 0 ＜ α ＜ 1,

K_actIt is measured value and the ratio for deducting the calculated value after the Model Self-Learning coefficient；

For additivity learning coefficient (such as temperature model), update according to the following formula：

$T_{i,j}^{new}=T_{i,j}+q_{i,j}{\mathrm{\αC}}_{act}---\left(15\right)$

Wherein α is smoothing factor,

C_actIt is the difference of the calculated value after measured value and the deduction Model Self-Learning coefficient,

When the actual corresponding operating point of rolling operating mode is closer to any one of characteristic point, then the corresponding power of the characteristic point Weight coefficient is relatively bigger, and it is bigger that it obtains the ratio for updating.

10. the self-learning method of a kind of operation of rolling Mathematical Modeling based on continuous curve surface according to claim 5, it is special Levy and be：

Wherein, after the completion of the belt steel rolling, the Model Self-Learning coefficient is updated.Actual rolling operating mode is calculated first (x_act,y_act,z_act),

Weight coefficient q in space in each described characteristic point_{I, j, k}：

Then the self study coefficient in renewal space in each characteristic point, updates according to the following formula for multiplying inquiry learning coefficient：

$T_{i,j,k}^{new}=T_{i,j,k}+q_{i,j,k}\mathrm{\α}(K_{act}-T_{i,j,k})---\left(17\right)$

Wherein α is smoothing factor, and span is 0 ＜ α ＜ 1,

K_actIt is measured value and the ratio for deducting the calculated value after the Model Self-Learning coefficient；

For additivity learning coefficient (such as temperature model), update according to the following formula：

$T_{i,j,k}^{new}=T_{i,j,k}+q_{i,j,k}{\mathrm{\αC}}_{act}---\left(18\right)$

Wherein α is smoothing factor,

C_actIt is the difference of the calculated value after measured value and the deduction Model Self-Learning coefficient,

When the actual corresponding operating point of rolling operating mode is closer to any one of characteristic point, then the corresponding power of the characteristic point Weight coefficient is relatively bigger, and it is bigger that it obtains the ratio for updating.